From 3cac3c32ff49618cfa05bb4e5630a62fc31af40a Mon Sep 17 00:00:00 2001
From: Gustaf Rydholm <gustaf.rydholm@gmail.com>
Date: Sun, 19 Jun 2022 19:55:30 +0200
Subject: Add stride and extra layer  to effnet

---
 .../networks/efficientnet/efficientnet.py          | 23 ++++++++++++++++++++--
 1 file changed, 21 insertions(+), 2 deletions(-)

(limited to 'text_recognizer/networks/efficientnet')

diff --git a/text_recognizer/networks/efficientnet/efficientnet.py b/text_recognizer/networks/efficientnet/efficientnet.py
index 2f2508d..de08457 100644
--- a/text_recognizer/networks/efficientnet/efficientnet.py
+++ b/text_recognizer/networks/efficientnet/efficientnet.py
@@ -36,6 +36,7 @@ class EfficientNet(nn.Module):
         bn_eps: float = 1.0e-3,
         depth: int = 7,
         out_channels: int = 1280,
+        stride: Tuple[int, int] = (2, 2),
     ) -> None:
         super().__init__()
         self.params = self._get_arch_params(arch)
@@ -43,6 +44,7 @@ class EfficientNet(nn.Module):
         self.bn_momentum = bn_momentum
         self.bn_eps = bn_eps
         self.depth = depth
+        self.stride = stride
         self.out_channels: int = out_channels
         self._conv_stem: nn.Sequential
         self._blocks: nn.ModuleList
@@ -66,7 +68,7 @@ class EfficientNet(nn.Module):
                 in_channels=in_channels,
                 out_channels=out_channels,
                 kernel_size=3,
-                stride=(2, 2),
+                stride=self.stride,
                 bias=False,
             ),
             nn.BatchNorm2d(
@@ -94,7 +96,24 @@ class EfficientNet(nn.Module):
         in_channels = round_filters(_block_args[-1].out_channels, self.params)
         self._conv_head = nn.Sequential(
             nn.Conv2d(
-                in_channels, self.out_channels, kernel_size=1, stride=1, bias=False
+                in_channels,
+                self.out_channels,
+                kernel_size=2,
+                stride=self.stride,
+                bias=False,
+            ),
+            nn.BatchNorm2d(
+                num_features=self.out_channels,
+                momentum=self.bn_momentum,
+                eps=self.bn_eps,
+            ),
+            nn.Mish(inplace=True),
+            nn.Conv2d(
+                self.out_channels,
+                self.out_channels,
+                kernel_size=1,
+                stride=1,
+                bias=False,
             ),
             nn.BatchNorm2d(
                 num_features=self.out_channels,
-- 
cgit v1.2.3-70-g09d2